A year with AI assistants in production: the experience of VTB, T-Bank, and Dodo Engineering

Developers from VTB, T-Bank, Dodo Engineering, and S7 TechLab shared their experience implementing AI coding assistants at the Conversations technical discussion, organized by Just AI, and reached unexpected conclusions about how roles in teams have changed over the past year.

What Changed in a Year of AI in Code?

Major Russian IT teams began massively implementing AI assistants for code writing about a year ago. Developer attitudes went through recognizable stages: enthusiasm at the start, disappointment when confronted with hallucinations and poor-quality code — and ultimately, calm adoption of the tool into daily routine.

The main question today is not "does AI work at all," but "has it changed anything measurable in team metrics?" According to the experience of discussion participants, there is real productivity improvement, but it's concentrated not where expected: not in the speed of writing new features, but in reducing labor costs on routine tasks — tests, documentation, refactoring of template code.

• Participating companies: VTB, T-Bank, Dodo Engineering, S7 TechLab
• Conversations discussion organizer: Just AI
• Key AI effect: savings on routine tasks, not acceleration of new feature development
• Tools have become established in the daily workflow of most teams

Should Developers Be Forced to Use AI?

One of the most acute questions of the discussion: should developers be required to use AI assistants? The experience of participants showed that the directive approach works worse than the organic one. VTB and T-Bank relied on internal advocates: developers who themselves demonstrated to colleagues cases of real time savings. When AI becomes "prescribed" through personal example, resistance decreases noticeably.

S7 TechLab and Dodo Engineering noted that negativity most often arises where the assistant generates code requiring lengthy review. If the ratio of "written by AI / accepted without significant changes" is too low, trust in the tool falls — and developers return to old methods.

"AI doesn't accelerate development automatically — it redistributes

where time goes."

Who Now Writes Code and Who Just Reviews It?

Here an unexpected shift in roles emerged. In teams where AI assistants are actively used, the developer's task shifts from writing code to reviewing and validating it. This changes requirements for juniors and mid-levels: the ability to quickly read, understand, and assess someone else's — including AI-generated — code becomes more important than the ability to write it quickly from scratch.

Participants identified two real risks. First: a developer accepts code "blindly" and accumulates technical debt. Second: spends so much time understanding the generated code that all the AI gains disappear. Both scenarios have already occurred in production teams.

What This Means

AI coding assistants have become established in the production cycle of major Russian IT teams — but along with efficiency, they have brought new questions about role structure, code review quality, and technical debt management. The main conclusion of the discussion: AI changes processes rather than simply accelerates them.

Frequently Asked Questions

Which companies participated in the discussion?

In the Conversations technical discussion organized by Just AI, representatives from VTB, T-Bank, Dodo Engineering, and S7 TechLab participated — all four companies shared their experience implementing AI assistants in production development.

Where is productivity from AI assistants actually growing?

According to participants' experience, productivity gains are most visible in routine tasks: writing tests, documentation, and refactoring template code — not in the speed of developing new features.